[boost] Re: Boost.Variant collaboration effort

From: Eric Friedman (ebf_at_[hidden])
Date: 2002-07-20 21:02:33

Keith Burton wrote:

[snip]
> 
> (3) why is there any need for thread-safety primitives ?
> All my experience tells me that putting 'thread safety' into classes 
> at this level just generates an illusion of safety with an unnecessary

> overhead for every use.
> 

O(1) complexity visitation can be guaranteed only (?) through a dispatch
table of some kind. That limits the options for how the dispatch table
object can be stored:

(1) Non-static member object.

This works, but it is inefficient as every instance of a particular
variant must create the table, which is the same for all variants of the
same type.

(2) Static member object.

This is problematic because it means a variant cannot be safely used in
other global static objects.

(3) Local static object.

Thread safety primitives are needed because the behavior of a local
static object is undefined in a multithreaded environment; accordingly,
thread-safety primitives are required so that access to the object is
serialized.

(See
http://www.cuj.com/articles/1999/9910/9910b/9910b.htm?topic=articles for
a short discussion of this problem in a different context.)

---
But, as Douglas Gregor noted, worrying about complexity guarantees for
variant visitation may be premature optimization.
So I guess cascading-if's (and O(N) complexity) are the way to go for
now. And maybe compiler optimizations will make the issue moot anyway.
Eric